Classification of human heart signals by novel feature extraction techniques for rescue application

In the paper, the methodology of detecting the heart rate signal wirelessly with regards to use it as an applications as an emerging tool for health care in emergency and surveillance application is presented. A novel feature extraction technique is presented to detect human heart signal or electrocardiogram (ECG) signals wirelessly and classify the signal as normal or abnormal. By the analysis of the heart rate signal the presence of an alive human person below the debris can be detected. After detection of heart rate signals via sensors, the challenge lies in classifying the human heart rate signals as normal or abnormal. Heart signals or Electrocardiogram (ECG) is the primary signal used to diagnose and detection of heart disease abnormality. In this work, ECG signals are de-noised and then analyzed to extract feature from it by a novel technique which is combining Discrete Wavelet Transform (DWT) and nonlinear, linear statistical feature. To verify the efficiency of extracted feature, here used various Machine Learning (ML) methods namely as Decision Trees (DT), Discriminant Analysis (DA), Logistic Regression (LR), Nearest Neighbor (KNN), Support Vector Machine (SVM) and Ensemble have been used as a classifiers and their results are compared. KNN reports the lowest accuracy of 81.60% on MIT- BIH Arrythmia Database. SVM and Ensemble reports the highest accuracy of 97.40%.